Best Room Response

[QMuse]

Well that's interesting, as the original post of mine that you objected to and which started this whole discussion was:

Well, I have done some more detailed measurements and it turned out that after 450Hz uncorrected response of both speakers converged and stayed converged to 20kHz. The small dip at 550Hz for which I thought was room driven showed at the other speaker response too and further measurements confirmed it is speaker's non-linearity.

As I said, the highest room mode I was able to confirm in my room is at 395Hz, 3db with Q=5, so I have to stand corrected.

 

[QMuse]

What I have also found is that, when using precise enough MMM measurement in the 20-900kHz range you can actually apply correction seamlessly on the reponse curve without caring much what is room and what is speaker. Of course, in reality in the 20-400Hz range you would be correcting room and in the 400-900Hz range it would be speaker. Interesting thing here is you can correct speaker pretty effectively in the 400-900Hz range with the MMM measurement, while doding that with gated sweep mesurement can present a challenge when done in room. From 900Hz upwards I would still, of course, recommend using gated sweep measruement.

 

[vavan]

I also subjectively prefer the results of wide band EQ based on these measurments

Even having praised f208s measured rather flat from 1m I also tend to prefer wide band eq as mmm in mlp (and my ears) detecting somewhat excessive energy on mid/highs and sometimes unexpected harshness that doesn't bother so much in planar headphones

 

[thewas]

The transition frequency depends on the room size and as said more controlled comparisons have to be done on what should be corrected or not.

When I measured with MMM 2 very good speakers (Dutch Dutch 8c & Neumann KH310) in a acoustically well treated room of a friend there was obviously a dominant room influence up to 500 Hz (if not even 1 kHz)

whereas in my smaller and acoustically worse room and with my less expensive speakers (Kali IN-8) counter-intuitively the obvious room influence is only up to approx. 300 Hz

Room acoustics still manage to surprise me everyday, so I try to stay open minded.

 

[Pio2001]

Hello,
Room modes are of course present all the way up to 20 kHz. You can see them on a sweep measurement with no smoothing :

The variations in high frequencies don't come from the speaker (the Klippel would see them), nor from the microphone. They come from the room.
The lowest mode in my room is 27 Hz. Which means that we might see up to 20000 / 27 = 741 occurences of it in my measurement.
In the low frequencies, the possible frequencies are at 27, 55, 82 Hz etc.
In higher frequencies, we can expect them at 10827, 10855, 10882 Hz etc. Since the scale is logarithmic, they are displayed closer to each other.
And there are also transversal and vertical modes, and also diagonal modes, and coupling with the next room etc.

Of course, if your perform an MM measurement, your microphone spends on average an equal amount of time inside peaks and inside notches, and they disappear from the graph.
In order to see all room modes on a graph, the microphone must not move during the measurement.

My main argument is that, at say 600Hz, the distance between the peak of a room mode and its null will be around 30cm.

If you use EQ to knock down the peak of that mode, this can only possibly be valid over a listening position range of less than 30cm.

Yes, IF what you measure comes from a room mode. But NOT if it comes from SBIR !
The smaller the speaker, the higher the SBIR of your front wall can go in frequency. The sound that the speaker sends backward at 600 Hz can bounce on the wall, interfere with the speaker (that's what is called SBIR), and the resulting interference goes through all the room unchanged, since all other reflections are lower in amplitude because of the distance.
We often talk about the fundamental modes of SBIR : a null around 100 Hz if the speaker is 80 cm from the wall, for example, but then, there is the peak at 200 Hz, then another null, and another peak, etc. These are much less dependent on the measurement position than the effects of room modes. They depend on the path difference between the direct sound and the indirect sound.

With a moving microphone measurement, you eliminate local room modes, but not SBIR.

I think the question is can you accurately enough measure a speakers FR response at the listening position separate from the room in order to make the "speaker adjustment". For speakers with published accurate on axis data you can just use that .... for the vast majority of speakers you would need to rely on measurements with REW or similar. As far as I know no one has verified that EQ'ing these type of in room measurements above Schroeder results in preferred sound quality.

One of the problem with speakers measurements from the listening position above the transition frequency is that you can't separate the effects of the direct sound from the effects of indirect sound. Therefore, if your speakers don't have a perfectly smooth directivity index, you can't know if you are correcting the direct on-axis sound of your speakers, or, on the opposite, spoiling it in order to correct their off-axis frequency response, that reaches the microphone later.
With an anechoic measurement, you can focus your correction on the listening window frequency response, which is impossible with a measurement from the listening position.

Another problem is the choice of the target curve, which should be different in each room, and for each speaker in a given room. Even with a speaker with a perfectly smooth directivity index, you can't know how much exactly the target curve at the listening position is tilted when the on-axis anechoic response of the speaker is flat.
Therefore any full band equalization based on measurements from the listening position is likely to miss the neutrality that you will get if you equalize directly the frequency response of the speaker alone.

And even when you have a neutral speaker, there is yet another problem, that I ran into while equalizing my system : while correcting SBIR and room modes (below 700 Hz in my case), where am I supposed to set the target ?

Here is my current room correction, obtained after more than 2 years of listening, with a dozen of intermediate trials :

I find it well balanced and it hasn't changed for more than one year now. The result looks like this :

Why is the slope higher in the 500 - 5000 Hz zone than in the 100 - 500 Hz zone ?
I don't know. It's just the only way it sounds right.
And why are low frequencies below 80 Hz quieter than the ones between 100 and 200 ? I don't know either. It just doesn't sound right if they are louder.

At frequencies where, from the listening position, we measure only room modes and SBIR, it is impossible to guess what would have been the ideal target curve if the room wouldn't have any modes and if there would have been no SBIR.
We can only guess... or listen and choose by ear.

 

[andreasmaaan]

@thewas_ could you explain how you took the MMM measurement in your friend's room?

And I'm curious what you/they decided to do (or not do) about the measured nonlinearities around 400-1000Hz?

Anyway, I'll go out on a limb in the meantime and give my 2c here now, but am willing to stand corrected if the facts turn out not to match the reality. But here goes:

When performing an MMM measurement, you would expect room nonlinearities to appear in the measured response right up to whatever frequency corresponds to the dimensions of the space through which you move the mic while performing the measurement (and indeed, if you hod the mic stationary, you would measure room modes right up to 20kHz+, so long as the room is reflective at those frequencies).

For example, let's say you move the mic through a sphere of diameter 35cm.

35cm is one wavelength at about 1000Hz. Therefore, in such a case, you would expect to measure a frequency response that was room-mode independent from 1000Hz up, with room modes beginning to show up between 1000Hz and 500Hz, and then dominating the measurement below 500Hz.

I would hazard a guess that the diameter of your MMM movement in your friend's room was about 35cm....?

Conversely, if you were to move the mic through a sphere of diameter 17.5cm, the same holds, but with the frequencies I just mentioned in the previous example doubled to 2000Hz and 1000Hz, respectively.

What I'm driving at is that what you're measuring here is not necessary telling you where the room becomes perceptually dominant, but rather how widely you swept the mic when performing the measurement.

 

[Pio2001]

here is a diagram showing how SBIR works :

But in reality, since we sit at equal distance from both speakers, it's more like this, and the result depends on the position, although not as strongly as if they were room modes :

 

[thewas]

Room modes are of course present all the way up to 20 kHz.

Of course, but above the transition frequency they overlap and thus do not dominate as so much but rather have a statistic character.

 

[thewas]

@thewas_ could you explain how you took the MMM measurement in your friend's room?

And I'm curious what you/they decided to do (or not do) about the measured nonlinearities around 400-1000Hz?

Anyway, I'll go out on a limb in the meantime and give my 2c here now, but am willing to stand corrected if the facts turn out not to match the reality. But here goes:

When performing an MMM measurement, you would expect room nonlinearities to appear in the measured response right up to whatever frequency corresponds to the dimensions of the space through which you move the mic while performing the measurement (and indeed, if you hod the mic stationary, you would measure room modes right up to 20kHz+, so long as the room is reflective at those frequencies).

For example, let's say you move the mic through a sphere of diameter 35cm.

35cm is one wavelength at about 1000Hz. Therefore, in such a case, you would expect to measure a frequency response that was room-mode independent from 1000Hz up, with room modes beginning to show up between 1000Hz and 500Hz, and then dominating the measurement below 500Hz.

If you move the mic through a sphere of diameter 17.5cm, the same holds, but with the frequencies I just mentioned in the previous example doubled to 2000Hz and 1000Hz, respectively.

What I'm driving at is that what you're measuring here is not necessary telling you where the room becomes perceptually dominant, but rather how widely you swept the mic when performing the measurement.

I usually move the microphone in a "cube" of approximately 60x60x30cm.
About such nonlinearities as above at 400-1000 Hz, I tend to correct them at my own setups, but for friends setups make usuall several different corrections and let them decide which they like most.

As you say the volume of the MMM sweep has also an influence on the region where individual modes start to become dominant, although my personal experience till now is that its smaller than I would expect.

 

[krabapple]

Is it possible that an un-optimized spectrum is the preferred response for your room? I've experimented with maybe 2 dozen custom targets and corresponding filters with Dirac.

I keep coming back to the un-optimized response being my preferred.

So theory...is it possible that your natural room's response will sound the most "real" with your system? Any alterations from the unaltered response seem to sound unnatural to me... Just a thought.

My room:

Of course it's possible but there's no reason to believe it would be generally true across listeners and rooms.

Recommendation from someone who has reviewed the psyhcoacoustic research in depth, Dr. Toole, is that correcting room issues (below Schroeder) is typically beneficial, while correcting speakers issues (above Schroeder) is typically beneficial *if* the speaker frequency response is deficient on or off axis in anaechoic measurements -- otherwise it can make things sound worse.

Individual personal preference, of course, trumps recommendations from statistics. Maybe you just like what you like, no matter what best practice says.

And maybe this should be a sort of 'sticky ' since such questions seem to arise frequently?

 

[RayDunzl]

Room modes are of course present all the way up to 20 kHz. You can see them on a sweep measurement with no smoothing :

The variations in high frequencies don't come from the speaker (the Klippel would see them), nor from the microphone.

Here's a JBL LSR 308 (red) vs Martin Logan reQuest (blue and foreground), at the listening position (same room)

 

[QMuse]

@thewas_ could you explain how you took the MMM measurement in your friend's room?

And I'm curious what you/they decided to do (or not do) about the measured nonlinearities around 400-1000Hz?

Anyway, I'll go out on a limb in the meantime and give my 2c here now, but am willing to stand corrected if the facts turn out not to match the reality. But here goes:

When performing an MMM measurement, you would expect room nonlinearities to appear in the measured response right up to whatever frequency corresponds to the dimensions of the space through which you move the mic while performing the measurement (and indeed, if you hod the mic stationary, you would measure room modes right up to 20kHz+, so long as the room is reflective at those frequencies).

For example, let's say you move the mic through a sphere of diameter 35cm.

35cm is one wavelength at about 1000Hz. Therefore, in such a case, you would expect to measure a frequency response that was room-mode independent from 1000Hz up, with room modes beginning to show up between 1000Hz and 500Hz, and then dominating the measurement below 500Hz.

I would hazard a guess that the diameter of your MMM movement in your friend's room was about 35cm....?

Conversely, if you were to move the mic through a sphere of diameter 17.5cm, the same holds, but with the frequencies I just mentioned in the previous example doubled to 2000Hz and 1000Hz, respectively.

What I'm driving at is that what you're measuring here is not necessary telling you where the room becomes perceptually dominant, but rather how widely you swept the mic when performing the measurement.

This is true, and for that reason even when doing single seat EQ I would recommend to cover app 80*50cm of horizontal area with 30-40cm of height. In this space at least 60 samples should be taken to if you mean to precisely EQ in the range up to 1000Hz.

I also suggest that speed of moving your mic is not higher than 5cm/sec. Not because of Doppler effect but to avoid sounds coming from cutting the air too fast. Using sponge protector is advised when doing MMM.

@andreasmaaan, @thewas_ I will try to repeat the experiment I did with introducing 650Hz resonance to the speaker's output. This time I would like to test if it would be possible to detect peak or dip at HF to check if MMM can be used for speaker's correction in this range. I was thinking about introducing +/-2dB at 5000Hz with Q=3. What are your thoughts, is this a good choice?

 

[QMuse]

Here's a JBL LSR 308 (red) vs Martin Logan reQuest (blue and foreground), at the listening position (same room)

View attachment 60437

Interesting! Can you plz repost this graph with 1/6 smoothing?

 

[RayDunzl]

Interesting! Can you plz repost this graph with 1/6 smoothing?

That exact picture was copied from another thread here.

This could be the source. If not, same thing, because it always is.

June 12 2017

JBL and MartinLogan uncorrected unsmoothed

Uncorrected 1/6 octave

Corrected unsmoothed:

Corrected 1/6 smoothing

 

[Pio2001]

Here's a JBL LSR 308 (red) vs Martin Logan reQuest (blue and foreground), at the listening position (same room)

Interesting !
I wonder if the absence of deep nulls in the Martin Logan is due to the very high directivity of the panel or to the fact that the diaphragm spans several deeps and null of any high frequency room mode.
I like the second explanation because we can see that the deeps in the raw frequency response diverge from 450 Hz and up. At that frequency, the number of deeps and the number of nulls across the height of the diaphragm begins to cancel out, while the directivity of the Martin Logan should be higher since 200 Hz (the crossover is at 180 Hz).

What I'd very much like to see is the RT60 measured values from the listening position with both speakers playing !
I mean that if the raw variation of the frequency response are smaller, there are chances that the measured RT60 curve is lower, because of the mathematical relationship between the micro-local slope of the frequency response (the slope of each micro-null) and the reverberation time in the room. The reverberation can be seen as the post-ringing of each of these hundreds of modes. If they are smaller, the reverberation might be shorter.

That would explain a lot of things about the sound of electrostatic panels ! It would also open the gate for more experiments : an electrostatic design would loose all its advantages in an anechoic room.

 

[Pio2001]

@andreasmaaan, @thewas_ I will try to repeat the experiment I did with introducing 650Hz resonance to the speaker's output. This time I would like to test if it would be possible to detect peak or dip at HF to check if MMM can be used for speaker's correction in this range. I was thinking about introducing +/-2dB at 5000Hz with Q=3. What are your thoughts, is this a good choice?

Obviously, if you introduce a change in the test signal playback, this change will be detected from the listening position.

The question are
1-how to know what target curve general slope gives a flat speaker response ?
2-how to distinguish a local frequency response variation caused by the on-axis speaker response (to be corrected) from the same variation caused by the off-axis speaker response only (not to be corrected) ?
3-how to know if local variations in the frequency response caused by the acoustics (variations in wall absorptions leading to irregular RT60 across the spectrum, SBIR...) should be corrected through increasing the total energy (filling the gaps), decreasing the total energy (removing resonances), or keeping the total energy constant ?... which relates to question 1, but here the question is about the general shape of the ideal target curve, rather than about the general slope.

For me, equalizing from the listening position is beneficial as long as the corrected problems are bigger than the uncertainty about these three questions. As soon as the problems to correct become smaller, equalizing from the listening position reaches its limits.

Equalizing the anechoic response of the speaker becomes necessary to improve the accuracy of the result relative to question 1.
Measuring the directivity index of the speaker (spinorama) becomes necessary in order to improve problems related to question 2, as far as possible (a bad directivity index can't be equalized).
And I'm not sure that there is any answer to question 3. It might well be a kind of "circle of confusion".

 

[RayDunzl]

I wonder if the absence of deep nulls in the Martin Logan is due to the very high directivity of the panel

That's my guess...

or to the fact that the diaphragm spans several deeps and null of any high frequency room mode.

Hadn't thought about that before. The panel is 15 x 48.

What I'd very much like to see is the RT60 measured values from the listening position with both speakers playing !

RT60

Uncorrected and Corrected

Something happened, can't redisplay T60M - data is grayed out now. Sometime old data may not have everything new versions of REW want to see.

---

OK...

Something new:

 

[QMuse]

Obviously, if you introduce a change in the test signal playback, this change will be detected from the listening position.

Only if measurement is accurate enough. And that is what is being tested here.

The question are
1-how to know what target curve general slope gives a flat speaker response ?
2-how to distinguish a local frequency response variation caused by the on-axis speaker response (to be corrected) from the same variation caused by the off-axis speaker response only (not to be corrected) ?
3-how to know if local variations in the frequency response caused by the acoustics (variations in wall absorptions leading to irregular RT60 across the spectrum, SBIR...) should be corrected through increasing the total energy (filling the gaps), decreasing the total energy (removing resonances), or keeping the total energy constant ?... which relates to question 1, but here the question is about the general shape of the ideal target curve, rather than about the general slope.

For me, equalizing from the listening position is beneficial as long as the corrected problems are bigger than the uncertainty about these three questions. As soon as the problems to correct become smaller, equalizing from the listening position reaches its limits.

Equalizing the anechoic response of the speaker becomes necessary to improve the accuracy of the result relative to question 1.
Measuring the directivity index of the speaker (spinorama) becomes necessary in order to improve problems related to question 2, as far as possible (a bad directivity index can't be equalized).
And I'm not sure that there is any answer to question 3. It might well be a kind of "circle of confusion".

Considering the very unusual shape of your pesonal target curve I'm not sure what "ideal target curve " means when coming from you.

 

[Pio2001]

RT60

Uncorrected and Corrected

Thank you for the data.

The difference is not spectacular.

 

[Pio2001]

Considering the very unusual shape of your pesonal target curve I'm not sure what "ideal target curve " means when coming from you.

Well, I prefer having a flat frequency response in the listening window of my speakers, calibrated by the manufacturer in anechoic conditions, than to try to blindly guess what may be close using measurements from the listening position.

I don't know if this is better. I think that this is more like "having the musicians playing in the room" while equalizing from the listening position would rather be like "being transported into the concert hall".